GPT Image 2.6 for UI design.
OpenAI released ChatGPT Images 2.6 on November 4, 2026, with two new API models, GPT-Image-2.6 Flare and Sunburst. For UI work, the useful changes are reference images, SVG export, animated GIFs, and markup-guided edits. Here is what changed and how to use it for app and website mockups.
Updated November 6, 2026 · 7 min read
What’s new in GPT Image 2.6
- Available in ChatGPT and Codex. OpenAI says Images 2.6 is available in ChatGPT and Codex across all tiers on web, desktop, and mobile.
- Reference images and GIFs. OpenAI’s launch adds animated GIFs and reference images to ChatGPT Images.
- Two API models.
gpt-image-2.6-flareis the smaller model for workflows where speed matters most.gpt-image-2.6-sunburstis the base model for more detail and precise edits. OpenAI says Flare can generate an image in as little as six seconds. - New output formats. Besides PNG, JPEG, and WebP, generations can return
svgandgif. Both models support transparent backgrounds. - Markup-guided edits. On an edit, you can upload a transparent overlay, for example a circle around one element, and describe the change.
- A complexity setting.
complexity(low,medium, orhigh) controls how much the model plans before rendering. Both models default tohigh.
Flare vs Sunburst
| Model | Best for | In UI work |
|---|---|---|
| GPT Image 2.6 Flare | Speed | Exploring many layout or style ideas quickly |
| GPT Image 2.6 Sunburst | Detail and precise edits | Final mockup images, small UI text, editing one element without changing the rest |
OpenAI’s migration advice: if you used GPT Image 2.5 Flare, start with 2.6 Flare; if you used 2.5 Sunburst, start with 2.6 Sunburst. If Sunburst meets your quality bar, try Flare with the same prompt and compare.
How to use it for UI design
- In ChatGPT: ask for the mockup in a chat and attach references if you have them.
- In Codex: describe the image, or include
$imagegento call the image generation skill explicitly. Attach a reference image to guide the result. - Through the API: call the images generations endpoint with
gpt-image-2.6-flareorgpt-image-2.6-sunburst.
OpenAI’s own guidance for interface previews: describe the product as if it already exists, focus on layout, hierarchy, spacing, and real interface elements, and avoid concept-art language so the result looks like a shipped interface rather than a sketch. Put any exact text in quotes.
Prompts for UI mockups
For 42 more, see the full GPT Image 2.6 prompts for UI design collection.
A mobile app screen
Create a realistic mobile app UI mockup for [a reading-habit app]. Show the home screen of a returning user: a header with a greeting, a large progress ring for today’s 20-minute goal, the current book with page progress, a primary button that says "Start a 6-minute session", and a bottom tab bar with Today, Library, Stats, Profile. Cream background, deep green actions, one coral accent, clear typography, minimal decoration. It should look like a real, shipped app. iPhone screen, no device frame.
A landing page hero in three directions
Create three desktop landing page hero sections for [an invoicing app for freelancers], same headline "Get paid 9 days faster." and the same two buttons, in three styles: warm editorial with a serif headline, bold and dark with one lime accent, soft and friendly with rounded cards. Realistic UI, exact text only, 16:9, no device frame.
A dashboard
Create a realistic web dashboard UI for [a coffee roastery]: left sidebar navigation, four KPI cards (active subscriptions 1,284; bags shipped 642; churn 2.1%; revenue $38.4k), a bar chart of weekly shipments for 8 weeks, and a list of this week's roasts with status tags. Calm, warm palette, clear hierarchy, real interface elements. 16:9.
An SVG icon set (API)
Set output_format="svg" and background="transparent" when you call the API, then use a prompt like:
A set of six simple line icons for an app's tab bar: home, library, stats, profile, search, settings. Consistent 2px stroke, rounded line caps, single color, no text, evenly spaced in one row.
A micro-interaction as a GIF
A short looping animation of a mobile button being tapped: the "Add to cart" button scales down slightly, turns green, and shows a check mark, then returns to its normal state. Flat UI style, white background, no extra text.
Edit one element with markup
Draw a circle around the element to change on a transparent overlay the same size as the image, then describe only that change:
Change only the circled button: make it the primary style with a solid dark green background and white label "Continue". Keep everything else in the screen exactly the same.
From image to a real screen
A mockup image is a picture: you cannot click it, resize it, or ship it. Once you have a direction you like, ask Codex to rebuild it as HTML:
Rebuild the attached mockup image as one self-contained HTML file at [390×844], with inline CSS and system fonts. Match the layout, spacing, colors, and text exactly. Draw icons and illustrations with HTML and CSS.
On a Codex Design canvas, the generated image and the HTML version sit side by side, so you can check the match and then edit fonts, colors, and spacing visually. For the full workflow, see how to make UI mockups with AI.
Frequently asked questions
Is GPT Image 2.6 free?
OpenAI says Images 2.6 is available in ChatGPT across all tiers, including on web, desktop, and mobile. Usage limits depend on your plan. API usage of Flare and Sunburst is billed per OpenAI’s API pricing.
GPT Image 2.6 vs 2.5: what should I use?
For a new UI workflow, start with 2.6. OpenAI maps 2.5 Flare to 2.6 Flare and 2.5 Sunburst to 2.6 Sunburst, and recommends comparing them with the same prompts, references, size, and quality.
Can GPT Image 2.6 make UI mockups?
Yes. OpenAI’s prompting guide includes interface previews as a use case. The results are images, so check small text and spacing, and rebuild the chosen direction as code when you want a real screen.